The SMF Works Project — Where AI Meets Humanity
← Back to self-hosting
General

Minimal configuration

  • GPU: NVIDIA RTX 3060 12 GB
  • RAM: 32 GB DDR4
  • Storage: 512 GB NVMe SSD
  • OS: Ubuntu 22.04 LTS
  • Use case: Running 7B parameter models locally via Ollama or llama.cpp
  • Recommended configuration

  • GPU: NVIDIA RTX 4090 24 GB or dual RTX 3090 24 GB
  • RAM: 64 GB DDR4/DDR5
  • Storage: 2 TB NVMe SSD
  • OS: Ubuntu 24.04 LTS
  • Use case: 13B–70B models, fine-tuning with LoRA, local ComfyUI/Stable Diffusion workflows
  • Premium configuration

  • GPU: NVIDIA A100 80 GB or H100
  • RAM: 128 GB ECC DDR5
  • Storage: 4 TB NVMe SSD
  • OS: Ubuntu 24.04 LTS or RHEL 9
  • Use case: Full fine-tuning, multi-user inference API, enterprise deployment